graphvision-ai 0.1.12

Automatic Graph Classification and Data Extraction

GraphVision AI 📊👁️

GraphVision AI is a lightweight, powerful computer vision library for automatic graph classification and structured data extraction.

Built with PyTorch and EasyOCR, it is designed to look at an image of a chart, instantly recognize what kind of graph it is, and extract its labels and values into a clean, developer-friendly JSON format.

---

✨ Key Features

🚀 Zero-Configuration

Models and weights are automatically downloaded from Hugging Face the first time you run it. No manual weight management required.

🧠 Intelligent Routing

Automatically classifies the input image (Pie, Vertical Bar, Horizontal Bar, Line, etc.) and routes it to the correct extraction algorithm.

🖼 Robust Input Handling

Pass a file path (String), an OpenCV image (NumPy array), or a PIL Image directly into the analyzer.

🔍 Smart OCR Masking

Uses contrast filtering and spatial mapping to accurately match text labels with their corresponding graphical data points.

---

📦 Installation

Install directly from PyPI:

pip install graphvision-ai

🚀 Quick Start

Extracting data from a graph takes less than 5 lines of code:

from graphvision.extractor import GraphExtractor
import json

try:
    vision_engine = GraphExtractor()
    
    # Path to your test image
    image_to_test = "hbar2.png" 
    
    # Run the extraction
    result = vision_engine.extract_data(image_to_test)
    
    print("\n✅ Extraction Successful!")
    print(json.dumps(result, indent=4))

except Exception as e:
    print(f"\n❌ Error during testing: {e}")

📄 Example Output

{
    "type": "pie",
    "title": "Favorite Programming Languages",
    "data": {
        "Python": 45.2,
        "JavaScript": 25.1,
        "C++": 15.4,
        "Java": 14.3
    }
}

📈 Supported Graph Types

Currently, GraphVision AI supports high-accuracy extraction for:

• pie — Pie Charts
• vbar_categorical — Vertical Bar Charts
• hbar_categorical — Horizontal Bar Charts

Line and Dot-Line charts coming soon.